Configuration Reference

The config.yaml file serves as a centralized configuration hub for the application, allowing users to specify various parameters and settings tailored to their requirements. It facilitates customization of the application’s behavior, including the selection and configuration of AI models. Users can modify the configuration file to adapt the application to different environments and use cases, enhancing flexibility and versatility.

Application Information

• name: The name of the application.

• version: The version of the application.

• description: A brief description of the application.

SDK Configuration

• sdk-tag: The version of the SDK to download.

Model Configuration

Note

When manipulating configured models, please refer to the Basics page for detailed instructions and best practices.

The models section allows users to specify AI models to be used in the application. Each model configuration includes the following parameters:

• name: The name of the model.

• path: The path of the directory that will contain the model.

• module: The Python module used for downloading the model.

• class: The Python class within the module.

• options: A map of additional options for configuring the model.

• tokenizers: List of tokenizer configurations.

  • path: The path of the directory that will contain the tokenizer.

  • class: The Python class used for downloading the tokenizer.

  • options: A map of additional options for configuring the tokenizer.

• pipelinetag: The pipeline tag used by the model identifies the specific task or functionality supported by the model (e.g., text-to-image, text-generation).

• source: The source of the model (e.g., hugging_face). In the case of a single file model, use custom source.

• addtobinaryfile: Indicates if the model should be added to the executable.

• isdownloaded: Indicates if the model has been downloaded.

• version: The version of the model.

Example

# Application information
name: "awesome-ia-app"
version: "1.0.0"
description: "Description for your awesome IA application."

# SDK Configuration
sdk-tag: "vX.Y.Z"

# Model Configuration
models:

    # Diffusers model
    - name: stabilityai/sdxl-turbo
        path: models/stabilityai/sdxl-turbo
        module: diffusers
        class: StableDiffusionXLPipeline
        options:
            torch_dtype: torch.float16
        tokenizers: []
        pipelinetag: text-to-image
        source: hugging_face
        addtobinaryfile: false
        isdownloaded: false
        version: "2023-12-07T18:04:49.000Z"

    # Transformers model
    - name: microsoft/phi-2
        path: models/microsoft/phi-2/model
        module: transformers
        class: PhiModel
        options:
            torch_dtype: '"auto"'
        tokenizers:
            - path: models/microsoft/phi-2/AutoTokenizer
                class: AutoTokenizer
                options:
                    do_lower_case: "True"
                    max_len: "128"
        pipelinetag: text-generation
        source: hugging_face
        addtobinaryfile: true
        isdownloaded: true
        version: "2024-02-06T12:36:24.000Z"

    # Single file model
    - name: stabilityai/sdxl-turbo
        path: models/sdxl-turbo.safetensors
        module: diffusers
        class: StableDiffusionXLPipeline
        options:
            torch_dtype: torch.float16
        tokenizers: []
        pipelinetag: text-to-image
        source: custom
        addtobinaryfile: true
        isdownloaded: true
        version: "2023-12-07T18:04:49.000Z"